Page 228 - Designing Autonomous Mobile Robots : Inside the Mindo f an Intellegent Machine

Becoming Unstuck in Time

Even given reasonably good dead reckoning and an appropriate fuzzy agent for the GPS,
there is going to be some amount of disagreement between the two systems. There-
fore, it may not be advisable to immediately accept every implied correction from the
GPS, because every time we accept a correction to the odometry we may invalidate
1
the data being compiled by the lidar agent .
Instead, we may wish to give the lidar time to report and be processed before accept-
ing any lateral or longitudinal correction from the GPS. By the time the lidar reports (or
runs out of time), the GPS may have two or even three implied corrections. We
don’t want to lose the richness of this data, but it is not necessary to use every reason-
able fix as it comes in.
Here we add a concept to the fuzzy navigation approach discussed in Chapter 12.
This factor is the inherent quality of each sensor for each degree of freedom. This
intrinsic quality is then multiplied by the quality of the fix for its respective degree of
freedom. By giving the lidar a higher intrinsic quality, we are telling the arbitration
logic that we trust the lidar more for heading and lateral position than we do the GPS.
Thus, our navigator task might be made to collect several implied corrections from
the GPS and at least one from the lidar before making a decision on which one to
believe. If we snapshot each of these implied corrections, and then compare their
qualities, we can give slower but more accurate sensors a chance to compete with
faster and less accurate sensors.

Data snapshots must include the full position and uncertainty estimates at the moment
of the data collection, along with any other information required. As long as no
other corrections have been made to the axis during the collection and processing
period, we can assume that we are still out of position by the same relative amount we
were when the data was collected. Therefore, when the relative correction to an axis is
calculated, this correction is applied to the present position estimate.

Issues of navigation data latency

There is also the issue of data latency. For example, if our GPS system uses a serial
interface to communicate its results to the robot, there will be a certain latency
between when the reading is made and when the navigation agent processes it. If

1 It is possible for an agent to translate incomplete data by the amount of the correction made by
another agent, but in practical terms I have found this to add noise to the data, especially in cases
where it occurs several times during collection.

211